src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 100     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 101     __m256i          vfitab;
 102     __m128i          vfitab_lo,vfitab_hi;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m256           dummy_mask,cutoff_mask;
 107     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 108     __m256           one     = _mm256_set1_ps(1.0);
 109     __m256           two     = _mm256_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm256_set1_ps(fr->ic->k_rf);
 124     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 125     crf              = _mm256_set1_ps(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 132
 133     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 134     rcutoff_scalar   = fr->rcoulomb;
 135     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 136     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144     j_coord_offsetE = 0;
 145     j_coord_offsetF = 0;
 146     j_coord_offsetG = 0;
 147     j_coord_offsetH = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 173
 174         fix0             = _mm256_setzero_ps();
 175         fiy0             = _mm256_setzero_ps();
 176         fiz0             = _mm256_setzero_ps();
 177
 178         /* Load parameters for i particles */
 179         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 180         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm256_setzero_ps();
 184         vvdwsum          = _mm256_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             jnrE             = jjnr[jidx+4];
 196             jnrF             = jjnr[jidx+5];
 197             jnrG             = jjnr[jidx+6];
 198             jnrH             = jjnr[jidx+7];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203             j_coord_offsetE  = DIM*jnrE;
 204             j_coord_offsetF  = DIM*jnrF;
 205             j_coord_offsetG  = DIM*jnrG;
 206             j_coord_offsetH  = DIM*jnrH;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 211                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 212                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 213                                                  &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm256_sub_ps(ix0,jx0);
 217             dy00             = _mm256_sub_ps(iy0,jy0);
 218             dz00             = _mm256_sub_ps(iz0,jz0);
 219
 220             /* Calculate squared distance and things based on it */
 221             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 222
 223             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 224
 225             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 229                                                                  charge+jnrC+0,charge+jnrD+0,
 230                                                                  charge+jnrE+0,charge+jnrF+0,
 231                                                                  charge+jnrG+0,charge+jnrH+0);
 232             vdwjidx0A        = 2*vdwtype[jnrA+0];
 233             vdwjidx0B        = 2*vdwtype[jnrB+0];
 234             vdwjidx0C        = 2*vdwtype[jnrC+0];
 235             vdwjidx0D        = 2*vdwtype[jnrD+0];
 236             vdwjidx0E        = 2*vdwtype[jnrE+0];
 237             vdwjidx0F        = 2*vdwtype[jnrF+0];
 238             vdwjidx0G        = 2*vdwtype[jnrG+0];
 239             vdwjidx0H        = 2*vdwtype[jnrH+0];
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 246             {
 247
 248             r00              = _mm256_mul_ps(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm256_mul_ps(iq0,jq0);
 252             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 253                                             vdwioffsetptr0+vdwjidx0B,
 254                                             vdwioffsetptr0+vdwjidx0C,
 255                                             vdwioffsetptr0+vdwjidx0D,
 256                                             vdwioffsetptr0+vdwjidx0E,
 257                                             vdwioffsetptr0+vdwjidx0F,
 258                                             vdwioffsetptr0+vdwjidx0G,
 259                                             vdwioffsetptr0+vdwjidx0H,
 260                                             &c6_00,&c12_00);
 261
 262             /* Calculate table index by multiplying r with table scale and truncate to integer */
 263             rt               = _mm256_mul_ps(r00,vftabscale);
 264             vfitab           = _mm256_cvttps_epi32(rt);
 265             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 266             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 267             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 268             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 269             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 270             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 271
 272             /* REACTION-FIELD ELECTROSTATICS */
 273             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 274             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 275
 276             /* CUBIC SPLINE TABLE DISPERSION */
 277             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 278                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 279             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 280                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 281             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 282                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 283             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 284                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 285             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 286             Heps             = _mm256_mul_ps(vfeps,H);
 287             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 288             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 289             vvdw6            = _mm256_mul_ps(c6_00,VV);
 290             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 291             fvdw6            = _mm256_mul_ps(c6_00,FF);
 292
 293             /* CUBIC SPLINE TABLE REPULSION */
 294             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 295             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 296             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 297                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 298             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 299                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 300             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 301                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 302             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 303                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 304             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 305             Heps             = _mm256_mul_ps(vfeps,H);
 306             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 307             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 308             vvdw12           = _mm256_mul_ps(c12_00,VV);
 309             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 310             fvdw12           = _mm256_mul_ps(c12_00,FF);
 311             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 312             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 313
 314             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velec            = _mm256_and_ps(velec,cutoff_mask);
 318             velecsum         = _mm256_add_ps(velecsum,velec);
 319             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 320             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 321
 322             fscal            = _mm256_add_ps(felec,fvdw);
 323
 324             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm256_mul_ps(fscal,dx00);
 328             ty               = _mm256_mul_ps(fscal,dy00);
 329             tz               = _mm256_mul_ps(fscal,dz00);
 330
 331             /* Update vectorial force */
 332             fix0             = _mm256_add_ps(fix0,tx);
 333             fiy0             = _mm256_add_ps(fiy0,ty);
 334             fiz0             = _mm256_add_ps(fiz0,tz);
 335
 336             fjptrA             = f+j_coord_offsetA;
 337             fjptrB             = f+j_coord_offsetB;
 338             fjptrC             = f+j_coord_offsetC;
 339             fjptrD             = f+j_coord_offsetD;
 340             fjptrE             = f+j_coord_offsetE;
 341             fjptrF             = f+j_coord_offsetF;
 342             fjptrG             = f+j_coord_offsetG;
 343             fjptrH             = f+j_coord_offsetH;
 344             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 345
 346             }
 347
 348             /* Inner loop uses 72 flops */
 349         }
 350
 351         if(jidx<j_index_end)
 352         {
 353
 354             /* Get j neighbor index, and coordinate index */
 355             jnrlistA         = jjnr[jidx];
 356             jnrlistB         = jjnr[jidx+1];
 357             jnrlistC         = jjnr[jidx+2];
 358             jnrlistD         = jjnr[jidx+3];
 359             jnrlistE         = jjnr[jidx+4];
 360             jnrlistF         = jjnr[jidx+5];
 361             jnrlistG         = jjnr[jidx+6];
 362             jnrlistH         = jjnr[jidx+7];
 363             /* Sign of each element will be negative for non-real atoms.
 364              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 365              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 366              */
 367             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 368                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 369
 370             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 371             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 372             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 373             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 374             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 375             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 376             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 377             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 378             j_coord_offsetA  = DIM*jnrA;
 379             j_coord_offsetB  = DIM*jnrB;
 380             j_coord_offsetC  = DIM*jnrC;
 381             j_coord_offsetD  = DIM*jnrD;
 382             j_coord_offsetE  = DIM*jnrE;
 383             j_coord_offsetF  = DIM*jnrF;
 384             j_coord_offsetG  = DIM*jnrG;
 385             j_coord_offsetH  = DIM*jnrH;
 386
 387             /* load j atom coordinates */
 388             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 389                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 390                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 391                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 392                                                  &jx0,&jy0,&jz0);
 393
 394             /* Calculate displacement vector */
 395             dx00             = _mm256_sub_ps(ix0,jx0);
 396             dy00             = _mm256_sub_ps(iy0,jy0);
 397             dz00             = _mm256_sub_ps(iz0,jz0);
 398
 399             /* Calculate squared distance and things based on it */
 400             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 401
 402             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 403
 404             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 405
 406             /* Load parameters for j particles */
 407             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 408                                                                  charge+jnrC+0,charge+jnrD+0,
 409                                                                  charge+jnrE+0,charge+jnrF+0,
 410                                                                  charge+jnrG+0,charge+jnrH+0);
 411             vdwjidx0A        = 2*vdwtype[jnrA+0];
 412             vdwjidx0B        = 2*vdwtype[jnrB+0];
 413             vdwjidx0C        = 2*vdwtype[jnrC+0];
 414             vdwjidx0D        = 2*vdwtype[jnrD+0];
 415             vdwjidx0E        = 2*vdwtype[jnrE+0];
 416             vdwjidx0F        = 2*vdwtype[jnrF+0];
 417             vdwjidx0G        = 2*vdwtype[jnrG+0];
 418             vdwjidx0H        = 2*vdwtype[jnrH+0];
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 425             {
 426
 427             r00              = _mm256_mul_ps(rsq00,rinv00);
 428             r00              = _mm256_andnot_ps(dummy_mask,r00);
 429
 430             /* Compute parameters for interactions between i and j atoms */
 431             qq00             = _mm256_mul_ps(iq0,jq0);
 432             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 433                                             vdwioffsetptr0+vdwjidx0B,
 434                                             vdwioffsetptr0+vdwjidx0C,
 435                                             vdwioffsetptr0+vdwjidx0D,
 436                                             vdwioffsetptr0+vdwjidx0E,
 437                                             vdwioffsetptr0+vdwjidx0F,
 438                                             vdwioffsetptr0+vdwjidx0G,
 439                                             vdwioffsetptr0+vdwjidx0H,
 440                                             &c6_00,&c12_00);
 441
 442             /* Calculate table index by multiplying r with table scale and truncate to integer */
 443             rt               = _mm256_mul_ps(r00,vftabscale);
 444             vfitab           = _mm256_cvttps_epi32(rt);
 445             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 446             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 447             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 448             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 449             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 450             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 451
 452             /* REACTION-FIELD ELECTROSTATICS */
 453             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 454             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 455
 456             /* CUBIC SPLINE TABLE DISPERSION */
 457             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 458                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 459             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 460                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 461             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 462                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 463             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 464                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 465             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 466             Heps             = _mm256_mul_ps(vfeps,H);
 467             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 468             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 469             vvdw6            = _mm256_mul_ps(c6_00,VV);
 470             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 471             fvdw6            = _mm256_mul_ps(c6_00,FF);
 472
 473             /* CUBIC SPLINE TABLE REPULSION */
 474             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 475             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 476             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 477                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 478             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 479                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 480             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 481                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 482             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 483                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 484             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 485             Heps             = _mm256_mul_ps(vfeps,H);
 486             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 487             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 488             vvdw12           = _mm256_mul_ps(c12_00,VV);
 489             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 490             fvdw12           = _mm256_mul_ps(c12_00,FF);
 491             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 492             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 493
 494             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 495
 496             /* Update potential sum for this i atom from the interaction with this j atom. */
 497             velec            = _mm256_and_ps(velec,cutoff_mask);
 498             velec            = _mm256_andnot_ps(dummy_mask,velec);
 499             velecsum         = _mm256_add_ps(velecsum,velec);
 500             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 501             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 502             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 503
 504             fscal            = _mm256_add_ps(felec,fvdw);
 505
 506             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 507
 508             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 509
 510             /* Calculate temporary vectorial force */
 511             tx               = _mm256_mul_ps(fscal,dx00);
 512             ty               = _mm256_mul_ps(fscal,dy00);
 513             tz               = _mm256_mul_ps(fscal,dz00);
 514
 515             /* Update vectorial force */
 516             fix0             = _mm256_add_ps(fix0,tx);
 517             fiy0             = _mm256_add_ps(fiy0,ty);
 518             fiz0             = _mm256_add_ps(fiz0,tz);
 519
 520             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 521             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 522             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 523             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 524             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 525             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 526             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 527             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 528             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 529
 530             }
 531
 532             /* Inner loop uses 73 flops */
 533         }
 534
 535         /* End of innermost loop */
 536
 537         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 538                                                  f+i_coord_offset,fshift+i_shift_offset);
 539
 540         ggid                        = gid[iidx];
 541         /* Update potential energies */
 542         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 543         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 544
 545         /* Increment number of inner iterations */
 546         inneriter                  += j_index_end - j_index_start;
 547
 548         /* Outer loop uses 9 flops */
 549     }
 550
 551     /* Increment number of outer iterations */
 552     outeriter        += nri;
 553
 554     /* Update outer/inner flops */
 555
 556     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 557 }
 558 /*
 559  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_256_single
 560  * Electrostatics interaction: ReactionField
 561  * VdW interaction:            CubicSplineTable
 562  * Geometry:                   Particle-Particle
 563  * Calculate force/pot:        Force
 564  */
 565 void
 566 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_256_single
 567                     (t_nblist                    * gmx_restrict       nlist,
 568                      rvec                        * gmx_restrict          xx,
 569                      rvec                        * gmx_restrict          ff,
 570                      t_forcerec                  * gmx_restrict          fr,
 571                      t_mdatoms                   * gmx_restrict     mdatoms,
 572                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 573                      t_nrnb                      * gmx_restrict        nrnb)
 574 {
 575     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 576      * just 0 for non-waters.
 577      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 578      * jnr indices corresponding to data put in the four positions in the SIMD register.
 579      */
 580     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 581     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 582     int              jnrA,jnrB,jnrC,jnrD;
 583     int              jnrE,jnrF,jnrG,jnrH;
 584     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 585     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 586     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 587     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 588     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 589     real             rcutoff_scalar;
 590     real             *shiftvec,*fshift,*x,*f;
 591     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 592     real             scratch[4*DIM];
 593     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 594     real *           vdwioffsetptr0;
 595     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 596     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 597     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 598     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 599     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 600     real             *charge;
 601     int              nvdwtype;
 602     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 603     int              *vdwtype;
 604     real             *vdwparam;
 605     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 606     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 607     __m256i          vfitab;
 608     __m128i          vfitab_lo,vfitab_hi;
 609     __m128i          ifour       = _mm_set1_epi32(4);
 610     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 611     real             *vftab;
 612     __m256           dummy_mask,cutoff_mask;
 613     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 614     __m256           one     = _mm256_set1_ps(1.0);
 615     __m256           two     = _mm256_set1_ps(2.0);
 616     x                = xx[0];
 617     f                = ff[0];
 618
 619     nri              = nlist->nri;
 620     iinr             = nlist->iinr;
 621     jindex           = nlist->jindex;
 622     jjnr             = nlist->jjnr;
 623     shiftidx         = nlist->shift;
 624     gid              = nlist->gid;
 625     shiftvec         = fr->shift_vec[0];
 626     fshift           = fr->fshift[0];
 627     facel            = _mm256_set1_ps(fr->epsfac);
 628     charge           = mdatoms->chargeA;
 629     krf              = _mm256_set1_ps(fr->ic->k_rf);
 630     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 631     crf              = _mm256_set1_ps(fr->ic->c_rf);
 632     nvdwtype         = fr->ntype;
 633     vdwparam         = fr->nbfp;
 634     vdwtype          = mdatoms->typeA;
 635
 636     vftab            = kernel_data->table_vdw->data;
 637     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 638
 639     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 640     rcutoff_scalar   = fr->rcoulomb;
 641     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 642     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 643
 644     /* Avoid stupid compiler warnings */
 645     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 646     j_coord_offsetA = 0;
 647     j_coord_offsetB = 0;
 648     j_coord_offsetC = 0;
 649     j_coord_offsetD = 0;
 650     j_coord_offsetE = 0;
 651     j_coord_offsetF = 0;
 652     j_coord_offsetG = 0;
 653     j_coord_offsetH = 0;
 654
 655     outeriter        = 0;
 656     inneriter        = 0;
 657
 658     for(iidx=0;iidx<4*DIM;iidx++)
 659     {
 660         scratch[iidx] = 0.0;
 661     }
 662
 663     /* Start outer loop over neighborlists */
 664     for(iidx=0; iidx<nri; iidx++)
 665     {
 666         /* Load shift vector for this list */
 667         i_shift_offset   = DIM*shiftidx[iidx];
 668
 669         /* Load limits for loop over neighbors */
 670         j_index_start    = jindex[iidx];
 671         j_index_end      = jindex[iidx+1];
 672
 673         /* Get outer coordinate index */
 674         inr              = iinr[iidx];
 675         i_coord_offset   = DIM*inr;
 676
 677         /* Load i particle coords and add shift vector */
 678         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 679
 680         fix0             = _mm256_setzero_ps();
 681         fiy0             = _mm256_setzero_ps();
 682         fiz0             = _mm256_setzero_ps();
 683
 684         /* Load parameters for i particles */
 685         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 686         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 687
 688         /* Start inner kernel loop */
 689         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 690         {
 691
 692             /* Get j neighbor index, and coordinate index */
 693             jnrA             = jjnr[jidx];
 694             jnrB             = jjnr[jidx+1];
 695             jnrC             = jjnr[jidx+2];
 696             jnrD             = jjnr[jidx+3];
 697             jnrE             = jjnr[jidx+4];
 698             jnrF             = jjnr[jidx+5];
 699             jnrG             = jjnr[jidx+6];
 700             jnrH             = jjnr[jidx+7];
 701             j_coord_offsetA  = DIM*jnrA;
 702             j_coord_offsetB  = DIM*jnrB;
 703             j_coord_offsetC  = DIM*jnrC;
 704             j_coord_offsetD  = DIM*jnrD;
 705             j_coord_offsetE  = DIM*jnrE;
 706             j_coord_offsetF  = DIM*jnrF;
 707             j_coord_offsetG  = DIM*jnrG;
 708             j_coord_offsetH  = DIM*jnrH;
 709
 710             /* load j atom coordinates */
 711             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 712                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 713                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 714                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 715                                                  &jx0,&jy0,&jz0);
 716
 717             /* Calculate displacement vector */
 718             dx00             = _mm256_sub_ps(ix0,jx0);
 719             dy00             = _mm256_sub_ps(iy0,jy0);
 720             dz00             = _mm256_sub_ps(iz0,jz0);
 721
 722             /* Calculate squared distance and things based on it */
 723             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 724
 725             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 726
 727             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 728
 729             /* Load parameters for j particles */
 730             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 731                                                                  charge+jnrC+0,charge+jnrD+0,
 732                                                                  charge+jnrE+0,charge+jnrF+0,
 733                                                                  charge+jnrG+0,charge+jnrH+0);
 734             vdwjidx0A        = 2*vdwtype[jnrA+0];
 735             vdwjidx0B        = 2*vdwtype[jnrB+0];
 736             vdwjidx0C        = 2*vdwtype[jnrC+0];
 737             vdwjidx0D        = 2*vdwtype[jnrD+0];
 738             vdwjidx0E        = 2*vdwtype[jnrE+0];
 739             vdwjidx0F        = 2*vdwtype[jnrF+0];
 740             vdwjidx0G        = 2*vdwtype[jnrG+0];
 741             vdwjidx0H        = 2*vdwtype[jnrH+0];
 742
 743             /**************************
 744              * CALCULATE INTERACTIONS *
 745              **************************/
 746
 747             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 748             {
 749
 750             r00              = _mm256_mul_ps(rsq00,rinv00);
 751
 752             /* Compute parameters for interactions between i and j atoms */
 753             qq00             = _mm256_mul_ps(iq0,jq0);
 754             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 755                                             vdwioffsetptr0+vdwjidx0B,
 756                                             vdwioffsetptr0+vdwjidx0C,
 757                                             vdwioffsetptr0+vdwjidx0D,
 758                                             vdwioffsetptr0+vdwjidx0E,
 759                                             vdwioffsetptr0+vdwjidx0F,
 760                                             vdwioffsetptr0+vdwjidx0G,
 761                                             vdwioffsetptr0+vdwjidx0H,
 762                                             &c6_00,&c12_00);
 763
 764             /* Calculate table index by multiplying r with table scale and truncate to integer */
 765             rt               = _mm256_mul_ps(r00,vftabscale);
 766             vfitab           = _mm256_cvttps_epi32(rt);
 767             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 768             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 769             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 770             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 771             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 772             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 773
 774             /* REACTION-FIELD ELECTROSTATICS */
 775             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 776
 777             /* CUBIC SPLINE TABLE DISPERSION */
 778             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 779                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 780             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 781                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 782             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 783                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 784             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 785                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 786             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 787             Heps             = _mm256_mul_ps(vfeps,H);
 788             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 789             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 790             fvdw6            = _mm256_mul_ps(c6_00,FF);
 791
 792             /* CUBIC SPLINE TABLE REPULSION */
 793             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 794             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 795             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 796                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 797             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 798                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 799             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 800                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 801             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 802                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 803             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 804             Heps             = _mm256_mul_ps(vfeps,H);
 805             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 806             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 807             fvdw12           = _mm256_mul_ps(c12_00,FF);
 808             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 809
 810             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 811
 812             fscal            = _mm256_add_ps(felec,fvdw);
 813
 814             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 815
 816             /* Calculate temporary vectorial force */
 817             tx               = _mm256_mul_ps(fscal,dx00);
 818             ty               = _mm256_mul_ps(fscal,dy00);
 819             tz               = _mm256_mul_ps(fscal,dz00);
 820
 821             /* Update vectorial force */
 822             fix0             = _mm256_add_ps(fix0,tx);
 823             fiy0             = _mm256_add_ps(fiy0,ty);
 824             fiz0             = _mm256_add_ps(fiz0,tz);
 825
 826             fjptrA             = f+j_coord_offsetA;
 827             fjptrB             = f+j_coord_offsetB;
 828             fjptrC             = f+j_coord_offsetC;
 829             fjptrD             = f+j_coord_offsetD;
 830             fjptrE             = f+j_coord_offsetE;
 831             fjptrF             = f+j_coord_offsetF;
 832             fjptrG             = f+j_coord_offsetG;
 833             fjptrH             = f+j_coord_offsetH;
 834             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 835
 836             }
 837
 838             /* Inner loop uses 57 flops */
 839         }
 840
 841         if(jidx<j_index_end)
 842         {
 843
 844             /* Get j neighbor index, and coordinate index */
 845             jnrlistA         = jjnr[jidx];
 846             jnrlistB         = jjnr[jidx+1];
 847             jnrlistC         = jjnr[jidx+2];
 848             jnrlistD         = jjnr[jidx+3];
 849             jnrlistE         = jjnr[jidx+4];
 850             jnrlistF         = jjnr[jidx+5];
 851             jnrlistG         = jjnr[jidx+6];
 852             jnrlistH         = jjnr[jidx+7];
 853             /* Sign of each element will be negative for non-real atoms.
 854              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 855              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 856              */
 857             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 858                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 859
 860             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 861             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 862             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 863             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 864             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 865             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 866             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 867             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 868             j_coord_offsetA  = DIM*jnrA;
 869             j_coord_offsetB  = DIM*jnrB;
 870             j_coord_offsetC  = DIM*jnrC;
 871             j_coord_offsetD  = DIM*jnrD;
 872             j_coord_offsetE  = DIM*jnrE;
 873             j_coord_offsetF  = DIM*jnrF;
 874             j_coord_offsetG  = DIM*jnrG;
 875             j_coord_offsetH  = DIM*jnrH;
 876
 877             /* load j atom coordinates */
 878             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 879                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 880                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 881                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 882                                                  &jx0,&jy0,&jz0);
 883
 884             /* Calculate displacement vector */
 885             dx00             = _mm256_sub_ps(ix0,jx0);
 886             dy00             = _mm256_sub_ps(iy0,jy0);
 887             dz00             = _mm256_sub_ps(iz0,jz0);
 888
 889             /* Calculate squared distance and things based on it */
 890             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 891
 892             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 893
 894             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 895
 896             /* Load parameters for j particles */
 897             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 898                                                                  charge+jnrC+0,charge+jnrD+0,
 899                                                                  charge+jnrE+0,charge+jnrF+0,
 900                                                                  charge+jnrG+0,charge+jnrH+0);
 901             vdwjidx0A        = 2*vdwtype[jnrA+0];
 902             vdwjidx0B        = 2*vdwtype[jnrB+0];
 903             vdwjidx0C        = 2*vdwtype[jnrC+0];
 904             vdwjidx0D        = 2*vdwtype[jnrD+0];
 905             vdwjidx0E        = 2*vdwtype[jnrE+0];
 906             vdwjidx0F        = 2*vdwtype[jnrF+0];
 907             vdwjidx0G        = 2*vdwtype[jnrG+0];
 908             vdwjidx0H        = 2*vdwtype[jnrH+0];
 909
 910             /**************************
 911              * CALCULATE INTERACTIONS *
 912              **************************/
 913
 914             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 915             {
 916
 917             r00              = _mm256_mul_ps(rsq00,rinv00);
 918             r00              = _mm256_andnot_ps(dummy_mask,r00);
 919
 920             /* Compute parameters for interactions between i and j atoms */
 921             qq00             = _mm256_mul_ps(iq0,jq0);
 922             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 923                                             vdwioffsetptr0+vdwjidx0B,
 924                                             vdwioffsetptr0+vdwjidx0C,
 925                                             vdwioffsetptr0+vdwjidx0D,
 926                                             vdwioffsetptr0+vdwjidx0E,
 927                                             vdwioffsetptr0+vdwjidx0F,
 928                                             vdwioffsetptr0+vdwjidx0G,
 929                                             vdwioffsetptr0+vdwjidx0H,
 930                                             &c6_00,&c12_00);
 931
 932             /* Calculate table index by multiplying r with table scale and truncate to integer */
 933             rt               = _mm256_mul_ps(r00,vftabscale);
 934             vfitab           = _mm256_cvttps_epi32(rt);
 935             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 936             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 937             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 938             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 939             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 940             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 941
 942             /* REACTION-FIELD ELECTROSTATICS */
 943             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 944
 945             /* CUBIC SPLINE TABLE DISPERSION */
 946             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 947                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 948             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 949                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 950             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 951                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 952             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 953                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 954             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 955             Heps             = _mm256_mul_ps(vfeps,H);
 956             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 957             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 958             fvdw6            = _mm256_mul_ps(c6_00,FF);
 959
 960             /* CUBIC SPLINE TABLE REPULSION */
 961             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 962             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 963             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 964                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 965             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 966                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 967             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 968                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 969             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 970                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 971             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 972             Heps             = _mm256_mul_ps(vfeps,H);
 973             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 974             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 975             fvdw12           = _mm256_mul_ps(c12_00,FF);
 976             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 977
 978             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 979
 980             fscal            = _mm256_add_ps(felec,fvdw);
 981
 982             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 983
 984             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 985
 986             /* Calculate temporary vectorial force */
 987             tx               = _mm256_mul_ps(fscal,dx00);
 988             ty               = _mm256_mul_ps(fscal,dy00);
 989             tz               = _mm256_mul_ps(fscal,dz00);
 990
 991             /* Update vectorial force */
 992             fix0             = _mm256_add_ps(fix0,tx);
 993             fiy0             = _mm256_add_ps(fiy0,ty);
 994             fiz0             = _mm256_add_ps(fiz0,tz);
 995
 996             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 997             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 998             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 999             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1000             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1001             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1002             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1003             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1004             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1005
1006             }
1007
1008             /* Inner loop uses 58 flops */
1009         }
1010
1011         /* End of innermost loop */
1012
1013         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1014                                                  f+i_coord_offset,fshift+i_shift_offset);
1015
1016         /* Increment number of inner iterations */
1017         inneriter                  += j_index_end - j_index_start;
1018
1019         /* Outer loop uses 7 flops */
1020     }
1021
1022     /* Increment number of outer iterations */
1023     outeriter        += nri;
1024
1025     /* Update outer/inner flops */
1026
1027     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*58);
1028 }